Fat-Tailed Dragon Asset Management (FTDAM)

FTDAM Partners:

Li-Chien Hsieh

Jinsoo Lee

Xiaowei Sun

S. John Suparman

Atsushi Yasutake

______

A Comparison of Screening-based Trading Strategies

We are comparing three screening-based trading strategies, which are:

Univariate screening method

Bivariate screening method

Univariate-based scoring method

We used the three different trading strategies to select various numbers of companies to buy and sell from the S&P 500. We tested our trading strategy from 1982 to 1999. The holding period is 1 year and the portfolio is reset at the end of March every year. We use the following two attributes as the basis for selection:

Trend in ROIC, defined as ROICt – ROIC t-1; ROIC= Operating Income/Invested Capital

Price Momentum, defined as (Pt – P t-1)/P t-1

Reasons for Attributes

We are trying to take advantage of both valuation-based and technical trading strategies. For starters, we chose one attribute each to catch the merits of those two strategies. For the valuation-based attribute, we chose change in ROIC. We believe the change in ROIC is a powerful leading indicator for the growth of revenue and earning, because invested capital works like a starter for company growth and the operating income gauges the effectiveness of the capital invested. For the technical attribute, we chose price momentum, because it is popular with successful tracking record.

Time Horizon

We chose one year as the holding period because we partially rely on valuation-based attribute (i.e., ROIC) and we believe that accounting based attribute can only work well for a horizon of at least one year. We tested those trading strategies over the period from 1982 to 1999. We reset the portfolio at the end of March every year, because we assume that all financial statements are available by that time.

Results

  1. Univariate – Top/Bottom 50 based on Price Momentum

We buy the top 50 companies and sell the bottom 50 ones based on stock price change of previous year. We get an average mean return of 8.88% with 23.68% volatility over the test period.

  1. Univariate- Top/Bottom based on Change in ROIC

We buy the top 50 companies and sell the bottom 50 ones based on the ranking of change in ROIC. We get an average mean return of 4.28% with 20.48% volatility over the test period.

  1. Univariate-based Top/Bottom 50 only (a scoring method)

Using the mean/variance of portfolios from the univariate screenings based on price momentum and change in ROIC, we optimize the weighting at a chosen volatility and get 79% for the price momentum portfolio and 21% for change in ROIC portfolio. Then we assign 79 and 21, respectively to the top 100 companies ranked by stock price change and change in ROIC. And we assign –79 and –21, respectively to the bottom 100 companies ranked by stock price change and change in ROIC. We sum up the scores and buy the 50 companies with top scores and sell the 50 companies with bottom scores. We obtain an average return of 10.85% and volatility of 24.45% over the test period.

  1. Univariate – Double Score

In this approach, we first assign score (negative score) to the 75 highest (75 lowest) ranking price momentum stocks and assign score (negative score) to the 150 highest (150 lowest) ranking ROIC stocks. We then add the two scores up and only buy stocks that are in the top 50 of price momentum ranking as well as in the top 50 of change in ROIC ranking. We sell stocks that are in the bottom 50 of both price momentum and change in ROIC rankings. The average return from this approach is higher at 12.77% and higher volatility of 28.69%.

  1. Univariate-based Quintile Scoring Method

We first rank stocks and divide them to quintiles by change in stock price and change in ROIC, respectively. We assign the scores of 5x10.88, 4x10.88, 3x10.88, 2x10.88, and 1x10.88, respectively for the quintiles of the ranking of stock price change. We assign 5x4.28, 4x4.28, 3x4.28, 2x4.28, and 1x4.28, respectively for the quintiles of the ranking of change in ROIC. The 10.88 comes from that the return is 10.88% using the univariate trading strategy based on price momentum, and 4.28 comes from that the return is 4.28% using the univariate trading strategy based on change in ROIC. Then we buy the 50 companies with top 50 total scores and sell the 50 companies with bottom total scores. We got an average return of 10.66% with volatility of 20.93% over the test period.

  1. Bivariate

We first sort on the ranking of change in stock price. Within the top and bottom 120 stocks, we further sort on change in ROIC. We buy the top/top 50 stocks and short the bottom/bottom 50 stocks. We got an average mean return of 9.78% with volatility of 22.65% over the test period.

Summary

Below is a summary table of the results of the last four trading strategies: (Excess return is the return minus the 30-year T-bond annual rate)

Top/Bottom 50 / Double Score / Bivariate / Quintile
Average Excess Return / 1.77% / 3.69% / 0.70% / 1.58%
Volatility of Excess Return / 24.45% / 28.69% / 22.65% / 20.93%

Conclusion and Further Improvements

After analyzing our results, we come to the conclusion that combining two attributes is better than just using one attribute. In a more general sense, sorting by two attributes results in a more refined results of winners. This finding emphasizes again the value of the scoring method applied in univariate screening, as more attributes can better forecast the stock returns and only scoring method can handle more attributes.

It is clear also from the summary table above that our four approaches actually yield similar results. One interesting pattern that emerges from this exercise is that the four approaches seem to work in the later years, i.e. giving positive excess returns, starting from the year 1994 up to 1998. This is because that our database is for the S&P 500 as in 2000, due to the data limitation. Therefore the closer to 2000, the more companies are in our database. If we could resolve this data limitation, we believe we could achieve even more promising results.

We think that further improvements can be made to the foundation work we have done here. For example, using more attributes, both accounting and technical, is expected to yield more promising results.